Classification Techniques in Machine Learning for Breast Cancer Prediction

Aashi Agarwal

breast cancer, classification models, logistic regression, support vector machine, decision tree, neural network.

Worldwide, female breast cancer is the most diagnosed cancer, with an estimated 2.3 million new cases (11.7%) every year. Breast cancer caused 685,000 deaths globally in 2020 [1].  Early detection is vital for effective treatment of breast cancer, and the survival rate for localized breast cancer is about 99% [2]. Screening mammograms are the most common way of detection, but still, 1 in every 8 mammogram misses to identify breast cancers. This paper focuses on using machine learning classification algorithms to help the diagnosis of breast cancer. The Wisconsin Breast Cancer Diagnostic dataset is used from the UCI machine learning repository. The dataset contains various characteristics of individual breast cancer cells obtained from a minimally invasive fine needle aspirate and classifies the cancer as Malignant or Benign. We train and test models for Logistic Regression, Logistic Regression with stochastic gradient descent, Support Vector Machine, Random Forest, Decision Tree, Boosted Decision Tree, and Neural Network with this dataset and compare their F1 score, false negatives, and false positives. Our objective is to create a model that has a very high recall, even at a slight expense of precision. A comparison is then made amongst the various techniques for the compute cost of training / inferencing and the performance of the model with respect to characteristics of the dataset. This paper then discusses how machine learning is being used currently for breast cancer prediction, and how we can make it better in the future to help early detection of breast cancer in every part of the world.

https://doi.org/10.31871/IJNTR.9.10.4